The Study On Optical Flow Of Color Image Based On Block Matching

Determining the motion and structure of objects in 3D scene by analyzing the motion information obtained from a sequence of time-varying 2D images, it is one of the problems that the actual research of computer vision focus on. One of the main tasks that the studying on time-varying image deals with, is vision motion estimation, and it's also the fundamental issue in computer vision. The research concentrates on how to extract the information of object such as structure, location and motion from 2D image sequences of scene, and further more, estimates the information of object such as 3D structure, location and motion. Because the optical flow field is provided with abundant information of motion and 3D structure, researching on optical flow field is always regarded as one of the effective way of solving the problem of vision motion analysis.In this thesis, we study the computation of the optical flow from real images and synthesized images especially the method based on block matching. It provides a essential theoretical base for the further work of getting object's structure information in 3D scene from optical flow field.At first, the definition, the theory and the computing method of optical flow are introduced. And the existing block motion searching methods are compared. Then the basic principle of color vision is referred in the next chapter, so we can apply somegray image theory to color one. With the fundamental knowledge above, a mixing searching method for block motion estimation is proposed to explore a novel approach for computing the optical flow of time-varying color image sequences. The highlight of our method is that it fully takes advantages of existing block matching algorithm, such as center-biased, midway stop, and motion vector prediction. And essential modifying is required to achieve a better efficiency. At last a new experimental method is presented and different scenes have been used to verify the algorithm.